netdev - CPU stuck due to the taprio hrtimer

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite for Android: free password hash cracker in your pocket

[<prev] [next>] [thread-next>] [day] [month] [year] [list]

Message-Id: <20240627055338.2186255-1-luyun@kylinos.cn>
Date: Thu, 27 Jun 2024 13:53:38 +0800
From: Yun Lu <luyun@...inos.cn>
To: vinicius.gomes@...el.com,
	jhs@...atatu.com,
	xiyou.wangcong@...il.com,
	jiri@...nulli.us
Cc: netdev@...r.kernel.org,
	linux-kernel@...r.kernel.org
Subject: CPU stuck due to the taprio hrtimer

Hello,

When I run a taprio test program on the latest kernel(v6.10-rc4), CPU stuck
is detected immediately, and the stack shows that CPU is stuck on taprio
hrtimer.

The reproducer program link:
https://github.com/xyyluyun/taprio_test/blob/main/taprio_test.c
gcc taprio_test.c -static -o taprio_test

In this program, start the taprio hrtimer which clockid is set to REALTIME, and
then adjust the system time by a significant value backwards. Thus, CPU will enter
an infinite loop in the__hrtimer_run_queues function, getting stuck and unable to
exit or respond to any interrupts.

I have tried to avoid this problem by apllying the following patch, and it does work.
But I am not sure if this can be the final solution?

Thanks.

Signed-off-by: Yun Lu <luyun@...inos.cn>
---
 net/sched/sch_taprio.c | 24 ++++++++++++++++++++++++
 1 file changed, 24 insertions(+)

diff --git a/net/sched/sch_taprio.c b/net/sched/sch_taprio.c
index a0d54b422186..2ff8d34bdbac 100644
--- a/net/sched/sch_taprio.c
+++ b/net/sched/sch_taprio.c
@@ -104,6 +104,7 @@ struct taprio_sched {
 	u32 max_sdu[TC_MAX_QUEUE]; /* save info from the user */
 	u32 fp[TC_QOPT_MAX_QUEUE]; /* only for dump and offloading */
 	u32 txtime_delay;
+	ktime_t offset;
 };
 
 struct __tc_taprio_qopt_offload {
@@ -170,6 +171,19 @@ static ktime_t sched_base_time(const struct sched_gate_list *sched)
 	return ns_to_ktime(sched->base_time);
 }
 
+static ktime_t taprio_get_offset(const struct taprio_sched *q)
+{
+	enum tk_offsets tk_offset = READ_ONCE(q->tk_offset);
+	ktime_t time = ktime_get();
+
+	switch (tk_offset) {
+	case TK_OFFS_MAX:
+		return 0;
+	default:
+		return ktime_sub_ns(ktime_mono_to_any(time, tk_offset), time);
+	}
+}
+
 static ktime_t taprio_mono_to_any(const struct taprio_sched *q, ktime_t mono)
 {
 	/* This pairs with WRITE_ONCE() in taprio_parse_clockid() */
@@ -918,6 +932,7 @@ static enum hrtimer_restart advance_sched(struct hrtimer *timer)
 	int num_tc = netdev_get_num_tc(dev);
 	struct sched_entry *entry, *next;
 	struct Qdisc *sch = q->root;
+	ktime_t now_offset = taprio_get_offset(q);
 	ktime_t end_time;
 	int tc;
 
@@ -957,6 +972,14 @@ static enum hrtimer_restart advance_sched(struct hrtimer *timer)
 	end_time = ktime_add_ns(entry->end_time, next->interval);
 	end_time = min_t(ktime_t, end_time, oper->cycle_end_time);
 
+	if (q->offset != now_offset) {
+		ktime_t diff = ktime_sub_ns(now_offset, q->offset);
+
+		end_time = ktime_add_ns(end_time, diff);
+		oper->cycle_end_time = ktime_add_ns(oper->cycle_end_time, diff);
+		q->offset = now_offset;
+	}
+
 	for (tc = 0; tc < num_tc; tc++) {
 		if (next->gate_duration[tc] == oper->cycle_time)
 			next->gate_close_time[tc] = KTIME_MAX;
@@ -1210,6 +1233,7 @@ static int taprio_get_start_time(struct Qdisc *sch,
 
 	base = sched_base_time(sched);
 	now = taprio_get_time(q);
+	q->offset = taprio_get_offset(q);
 
 	if (ktime_after(base, now)) {
 		*start = base;
-- 
2.34.1